[ClusterLabs] Antw: Re: Antw: [EXT] what is the "best" way to completely shutdown a two‑node cluster ?

[Ulrich Windl]

>>> "Lentes, Bernd" <bernd.lentes at helmholtz-muenchen.de> schrieb am 07.02.2022 um
14:52 in Nachricht
<1860067605.168352603.1644241977151.JavaMail.zimbra at helmholtz-muenchen.de>:

> 
> ----- On Feb 7, 2022, at 2:36 PM, Ulrich Windl 
> Ulrich.Windl at rz.uni-regensburg.de wrote:
> 
> 
>> 
>> Bernd,
>> 
>> what if you set the node affected to standby, or shut down the cluster
>> services? Or all all nodes powered by the same UPS?
> 
> All nodes are powered by the same UPS.
> 
>> 
>> 
>>> 
>>> And what is if both nodes are running ? Can i do that simultaneously on both
>> 
>>> nodes ?
>> 
>> I guess that should work.
>> 
>>> My OS is SLES 12 SP5, pacemaker is 1.1.23, corosync is 2.3.6-9.13.1
>> 
>> Your action plan depends on what the VMNs are doing: basically every HA
>> resource should survive a hard restart without much damange.
> 
> Well, some vm's have databases, i'd like to shutdown these cleanly.
> 
>> So maybe an option could be: Do nothing, or do an emergency shutdown of the
>> node without properly migrating all the VMs elsewhere.
> 
> The VM's don't need to be migrated, the whole cluster should stop in a 
> reasonable time and manner.
> 
>> You cannot make an application HA by putting it in a VM; at least not in
>> general.
> 
> I know. But the time gap between one node having problems and booting the vm 
> on the other node
> is ok for us.

So my proposal would be:
Assign priorities to your VMs (resources) and start to shutdown all nodes cleanly something like 10 minutes before the estimated end of battery power.
Hopefully all will be down, unmounted and synced when the power goes down then.

Regards,
Ulrich


> 
> Bernd